Re: Planner problems in 8.2.4 and 8.2.5 (was: Possible planner bug/regression introduced in 8.2.5)

"Greg Sabino Mullane" <greg(at)turnstep(dot)com> writes:
> Can we switch the default_statistics_target to a default of 100? Is
> there any reason not to do so at this point?

Other than a 10x increase in the cost of ANALYZE, and in the cost of
histogram-based operations in the planner?

It's been clear for quite awhile that a stats target of 10 is often
too low, but no one has done the legwork to establish what a more
reasonable tradeoff point would be. I'm not for increasing the overhead
by a factor of 10 ... or even a factor of 2 ... without some real
evidence about the implications.

I'd be inclined to approach it by first trying to establish what's a
sane default for operations unrelated to LIKE estimation. If that comes
out as 50 or more, I'd just lower the LIKE threshold to whatever it
comes out at. The 100 number was really just a guess in the first
place.

regards, tom lane